Nemotron-RL-Identity-Following-v1-prompt-only
收藏资源简介:
Nemotron-RL-Identity-Following-v1-prompt-only 是一个从源数据集 nvidia/Nemotron-RL-Identity-Following-v1 中提取的仅包含提示(prompt)的数据集。该数据集旨在为强化学习身份遵循任务提供提示数据,适用于模型后训练、提示工程或微调等场景。数据内容主要包括一个CSV文件(prompts.csv),其中每行对应一个提取的提示记录,包含prompt、分离的system_prompt以及当源行定义了可用工具时的结构化tools(嵌套值以JSON编码存储)。数据集规模为21660条提取行,无失败提示行,行数差异为零。此外,数据集还提供了摘要文件(summary.md)和空值行索引文件(null_or_empty_rows.md),以辅助数据质量检查。数据集标签包括nemotron、prompt-only和post-training,表明其与Nemotron项目、纯提示数据及后训练流程相关。
Nemotron-RL-Identity-Following-v1-prompt-only is a dataset extracted solely from the prompts of the source dataset nvidia/Nemotron-RL-Identity-Following-v1. It aims to provide prompt data for reinforcement learning identity following tasks, suitable for scenarios such as model post-training, prompt engineering, or fine-tuning. The data primarily consists of a CSV file (prompts.csv), where each row corresponds to an extracted prompt record, containing the prompt, a separated system_prompt, and structured tools (with nested values stored in JSON encoding) when the source row defines available tools. The dataset scale is 21,660 extracted rows, with no failed prompt rows and a row count difference of zero. Additionally, the dataset includes summary files (summary.md) and null or empty row index files (null_or_empty_rows.md) to assist with data quality checks. The dataset tags include nemotron, prompt-only, and post-training, indicating its relevance to the Nemotron project, pure prompt data, and post-training workflows.




